Again and again
When you have worked with Solaris 11 or 10 you have surely used SMF and loved it to a varying degree. Since 11.3 there is something new in it: SMF knows now a concept called “periodic and scheduled services”.
What is special about those new kinds of services? Well quite simple.Periodic services are services that are running from the start of the service until the system stops every soandso seconds. For example run the method defined in service periodic/dbcheck
every 60 seconds.Scheduled services are running from the start of the service until the system stops in the configured schedule. For example run the method defined in service scheduled/dbbackup
daily at 01:00 am.
Sounds familiar. Well, yeah. You can think about it as a cron on steroids. But there are some reasons why this is better than just cron.
You can use all advantages of SMF for example like the simple delegation of administrative tasks to other users. You can the pkg format to simply add an action to a package that add such an service.
For me even more important is that periodic services knows the concept of jitter. This is quite useful: Simple cron has a slight disadvantage. Imagine you have for example several zones zones with a process scheduled to start at 01:00. In this case cron will start them at the same time.
You get a formidable thundering herd of processes. Despite the fact that system are quite powerful today, wouldn’t it be nice that not all processes start at the same time but at point in time with a slight difference. The mentioned parameter jitter
enables such a slight difference. On the other side there is a parameter called delay that configures what the delay of the start after the service got into the online state.
Another interesting capability is the ability to set a parameter called recover. In case your system was down for a while and server method should have run during the downtime, the service will be started during startup in case this parameter is set to true.
Preparations
In order to demonstrate this i have to prepare a few things first. We need a method-script. I will just use a short script that is doing nothing except saying that it does nothing.
We will use this script later on in the examples.
Periodic services
Okay, lets try a periodic service at first. Create a XML manifest for it.
Based on this manifest file the script runs every 60 to 65 seconds and the first time after 5 seconds after the service got online To install this service, i will use the manifest-import.
Due to the enabled='false'
in the manifest it’s initially disabled.
So we enable the service no
Okay, the periodic services know two parameters. NRUN
and LRUN
. NRUN
tells you when the service will run the next time. LRUN
tells you when the service did run the last time.
As you see the service is not running in perfect 60 interfaces but in the configured interval with some jitter in it.
Scheduled services
Okay, now lets do the same for a scheduled service.
Now we have to install and enable this service as well.
When we look at the NRUN value, we will see that this service will run 12:40:44 (in reality it was around 8 o’clock in the evening … this happens without resetting the time in a Virtualbox)
As we have just created this schedules service, obviously there is no last run-date so far.
However when we wait until the configured time and ask for the last run timestamp, we will see a value here.
As indicated by the NRUN value, the service ran at 12:40:44 and so is now the the LRUN value.
List of scheduled services
It’s quite easy to find out, which scheduled services are active in the system.
An alternative way to do so, would be checking which services are dependend to the restarter svc:/system/svc/periodic-restarter:default
. This is the delegated restarter used to do the scheduled and periodic restart.